System Level Diagnosis: A Review

System-level diagnosis appears to be a viable alternative to circuit-level testing in a complex multiprocessor system. System-level diagnosis is also a very useful tool in analyzing system fault detection and location capabilities of a complex multiprocessor system. The conventional t-diagnosability theory characterizes a system to be uniquely diagnosable under the presence of any arbitrary fault set of size up to t. It assumes the diagnosis process is centralized and all the outcomes of tests are available before the diagnosis starts. It, however, does not consider the testing and reconnguration mechanisms involved in performing this task. The conventional theory also fails to answer the level of diagnosability if a system is not completely diagnosable with respect to all possible fault sets of up to certain size. These questions has been raised and and solutions have been found recently by several researcher. Distributed diagnosis algorithms, taking advantage of inherent parallelism in these systems, have also been developed. This paper surveys all the new results which are expected to aaect the fault diagnosis methods in future multiprocessor systems.